[SOLVED] Arm based rival to your x86 home pc?

jefro · 12-03-2020, 03:05 PM

I've seen that big companies buy smaller ones for what "they" need. Cheaper to buy than to negotiate what few items they want. Then the rest is either left to rot or sold off.

They want or need something in that deal.

zeebra · 12-04-2020, 07:34 AM

Quote:

Originally Posted by business_kid

Background: It appears Apple are switching processors yet again from X86 to Arm, which seems to be a vote of confidence in Arm. I hardly expect Apple CPUs or SBCs to come on the market. But Apple obviously sees the potential in Arm.

Question: Now the software is there to run linux on Arm, but is the hardware? What's the best offering? Is it the RazPi 4?

Well, you need to be concerned about the software as well. Arm is niche in the GNU/Linux world, although alot is available, alot is NOT cross-compiled/ported to arm, so you might need to cross compile your favourite software XYZ to your arm setup manually as well. Overall your software and distro options are greatly reduced if you switch to arm.

business_kid · 12-04-2020, 08:48 AM

Quote:

Originally Posted by obobskivich

My SWAG as to why nVidia did this is for a few reasons:
1) They're 'over-valued' right now and probably trying to get something more tangible out of that 'equity' before it corrects down.
2) If I remember right, they actually integrate ARM chips into their GPUs as supervisor controllers, I read somewhere they had been looking at RISC-V to replace it (because the 'current' solution is supposedly starting to be inadequate vs the complexity of their newer GPUs), but probably it made more sense to someone to keep everything more 'locked down.' If I remember right the lack of documentation/support for that supervisor IC is also why noveau performance takes such a hit vs their proprietary driver.
3) They've been shipping the exact kind of ARM + GeForce APU you describe for a while, with Jetson and Tegra (as well as some mobile chipsets for Nintendo), and they push this hardware for machine vision, IoT, machine learning, etc. If I remember right the 'big market' for them is self-driving/autonomous vehicles and they're probably trying to better cement their position as that segment starts to expand.
4) They probably also have some fantasy of re-capturing the more mainstream gaming market (consoles/arcades/etc which is a multi-billion/yr industry) from AMD by being able to offer a complete APU package (and/or getting more into 'mobile' against Qualcomm and Samsung), but A) they've already been trying that (since they were/are an ARM-licensee) and it hasn't taken off and B) it would require aarch64 to be more supported by game publishers (Microsoft actually does support this 'in house' - most of the Snapdragon laptops I linked ship with Windows 10 out of the box for example, but good luck throwing games on them), which hasn't materialized (again apart from Nintendo's closed platform).

$4×10^10?? $40 Billion??

Frankly, I don't see any of those as reasons to make such a major investment. There is another reason why software houses are bought, and that is for the talent - the engineers. Those acquisitions are usually under $20 Million, because talent is mobile.

There's a big expansion on here at Nvidia, or a long term plan. AI, machine learning, or self-driving cars might be it. But at $40 Billion, Arm must be pivotal, or currently it's grossly undervalued. We can guess after they acquire the next company.

And I don't think software for Arm is a big issue any more. Many maintainers have already pushed patches to make their code compile on Arm, so it's easier sailing now. Perl, Java & Python are all there, and even wine is running some x86 code. Sure, you'll never get Flash on Arm, unless Adobe hires and manages to keep a real programmer. But who wants that?

obobskivich · 12-04-2020, 08:53 AM

Quote:

Originally Posted by business_kid

$4×10^10?? $40 Billion??

Frankly, I don't see any of those as reasons to make such a major investment. There is another reason why software houses are bought, and that is for the talent - the engineers. There's a big expansion on here. AI, machine learning, or self-driving might be it. But at $40 Billion, Arm must be pivotal, or grossly undervalued. Those acquisitions are usually under $20 Million, because talent is mobile.

This is a rational and coherent analysis, but we're currently living through another tech bubble, so of course everything is going to be inflated. I'm also guessing that 'acquihire' wouldn't quite work here, because there's patents and licences that can't be so easily moved, and nVidia probably feared where ARM might otherwise end up (Softbank unloaded it to try and stay afloat, if you believe the pundits, so *someone* was going to buy it).

business_kid · 12-04-2020, 12:58 PM

Well, if you look at nvidia.com, you can see what Nvidia are trying to hawk.

They have a Geforce on Apple Mobiles and IPADs.
Some ndx or gdx workstation for AI
Some GPU for supercomputing
Some infiniband controller
Requests for submissions to some posh sounding conference

Nvidia has "More" buttons on all of them, but I'm not pressing them, because of my low tolerance of sales blurb and 'forced wait' in webpages. Jefro posted an interesting link about servers, and I missed it until now It appears AMD are getting in on the Arm server act. So AMD can see the future with X86 is going to plateau. They released an Opteron which is Arm64 based, so they are at least dipping a toe into the water there.

So Nvidia bought Arm as the next big thing in computing after X86_64. It comes with it's existing mobile business, it's Industrial SoC business & tablet business. Arm has potential anywhere electronics is made, up to AI, super computing, high speed comms, etc. Software will catch up, and in many cases already has. X86_64 is power hungry. It has to be, because it's little endian. Arm simply has the potential to get the most computing power out of any given silicon wafer.

business_kid · 12-06-2020, 09:30 AM

Returning vaguely to post #1, I finally found one: https://store.avantek.co.uk/workstat...rice=1000-2000

That's a 32 core Arm server on the ThunderX cpu - the ThunderX is about a 5 year old design at this stage. If you strip off the '?' and everything after on the above link, you'll see the range.This is the cheapest I could find. These are marked as workstations for research & development only. The ThunderX2 CPU, which is at the cheap server races comes for about $13K(+ the rest). There's a vastness of I/O potential that I certainly am not paying for. You are supposed to populate the beast with a few hundred GB of ram at least, and many TB of HDD/SSD/NVMe, and graphics is an afterthought. But at least, you can get out your debit card and buy one.

All the rest seem to be broken links, failed businesses or projects, or seriously expensive alternatives aimed at Volume OEMs. Then there's a huge gap down to the likes of the RazPi - the pipsqueak class of Arm we are probably familiar with.

It seems the best way to get one of these interesting things is to know somebody at the right companies or get hands on a sample product some big company decided not to buy.